i think it would take too much ram to be done, unless some kind of sd card streaming, but i have no idea of the feasibility.
About the Fourier transform, you're talking about some kind of resynthesis which, i guess, would need lot of ram and cpu.
I feel that we would benefit more from some sound design utilities.
If you wish to investigate that way, i can suggest some work i done about sample mimiking with fm :
https://github.com/pvig/preen-mutatorit's a puredata project which aim is to analyze an audio sample to get some metrics from it, and then generate random preenfm patch, to begin with,
then try to mutate it to make it converge to the audio sample.
(inspired from
https://www.jstor.org/stable/3680541)
Well, it does not work very well, it's not really better than the random patch feature of the preenfm
The good thing with this approach is that it would not need more cpu usage on the preenfm.
Maybe someone can pick up this idea an try to do something out of it ?
About the current features, here is another fw version with optimizations and slightly enhanced granular fx, especially Grain2 can now be tuned to the octave.